Inconsistency in Panther vs KOBAS2.0. Other recommendation for KEGG?

Question

[CandiceChuDVM]

Please see my latest answer for the inconsistency in Panther vs KOBAS2.0.

Can anyone recommend other platforms for KEGG?

==================================================================

Hi all,

I am analyzing RNA-Seq data from dogs. I would like to see if my differentially expressed gene (DE genes) are enriched in specific GO terms or certain pathways. However, I don't know if I should go with Panther or KEGG.

I noticed that there are 19714 annotated canine genes in Panther. Among them, 2509 genes are mapped to 165 pathways. However, in KEGG, there are 24827 genes and 298 pathways.

Does that mean I should use KEGG since it has more genes/pathways in its database?
And, what are the differences between Panther and KEGG?

Answer 1

[Benn]

Why don't you try both?

I use GO or pathway enrichment analysis to get (biological) insight from my data. So for me it's not a question which one is better or more suitable, but just try both and see which one gives you most insight to help you understand your experiment.

[WouterDeCoster]

More is interesting to get a broader insight, but also leads to cherry picking which gives you something matching your hypothesis.

[Benn]

Yes, I agree with your statement @decosterwouter and that's the case for all research in science (not only bio).

However, most RNA-seq (and other high-throughput) analyses are not hypothesis driven, but hypothesis generating... At least they should be ;-P

Answer 2

[natasha.sernova]

There were some posts:

Best Way To Do Pathway Analysis Of A Set Of Genes?

A: Can someone give insight into the difference between current gene-function datab

But I haven't seen the statement which one is better, Panther or KEGG.

Panther site: http://www.pantherdb.org/about.jsp

KEGG looks more fundamental, it has more genomes according to my impression.

http://www.genome.jp/kegg/catalog/org_list.html

On the other hand there is a very fresh Panther update (below):

http://www.ncbi.nlm.nih.gov/pubmed/26578592

I would say: the choice depends upon what you need.

GO Enrichment Analysis

http://geneontology.org/page/go-enrichment-analysis

It is Panther without any saying.

As far I have understood, you would like to see

if your differentially expressed gene (DE genes) are enriched in specific

GO terms or certain pathways. I would choose Panther.

KEGG may do a lot of things http://www.genome.jp/en/gn_tools.html

and also some others, http://www.genome.jp/en/gn_kubic.html

but not gene enrichment, am I right?

I've compared the number of articles in NCBI:

http://www.ncbi.nlm.nih.gov/pubmed/?term=KEGG+AND+GE+AND+dog - 8 papers

http://www.ncbi.nlm.nih.gov/pubmed/?term=Panther+AND+GE+AND+dog - 29 papers

So - Panther?

Answer 3

[CandiceChuDVM]

Hi all,

I used both KOBAS 2.0 (an update of KOBAS: KEGG Orthology Based Annotation System) and Panther with binomial test to analyze my data with 1158 DE genes.

Unsurprisingly, they gave me different results:

In Panther, 1100 genes were accepted. Integrin signalling pathway (P00034) is the most prominent one with p-value = 1.42E-05 followed by ATP sythesis (P02721) with p-value = 4.37E-02.

In KOBAS 2.0, 764 genes were accepted. ECM receptor interaction (cfa 04512) is the most significant one with a p-value = 7.25E-06. However, ATP sythesis (P02721) also appears in KOBAS 2.0 and has the smallest p-value (0.016) comparing to other Panther pathways that exists in KOBAS 2.0. On the other hand, p-value of Integrin signalling pathway (P00034) is 0.071.

I did a little research to find out why:

1. I noticed that "KOBAS 2.0 incorporates knowledge across over 2969 species from 6 pathway databases (KEGG PATHWAY, PID, BioCarta (from PID), Reactome, BioCyc, and PANTHER)". Despite that KOBAS incorporate Panther, KOBAS has only 41 out of 165 Panther pathways.

2. In KOBAS 2.0, you can change your ideal statistical methods (hypergeometric test / Fisher's exact test, binomial test, and chi-square test), small term cutoff, and perform FDR corrections (reference1). Panther only use binomial test (reference2) and there is no small term cutoff and FDR correction (please correct me if I am wrong).

3. As you can see, I use the same statistical method but got different results. It is caused by the fact that there are different numbers of genes in the Panther pathways that exist in KOBAS 2.0. For ATP sythesis (P02721), Panther shows there are 10 genes but KOBAS 2.0 shows there are only 2 genes.

I don't like to do cherry picking in research but it is obviously difficult for me to decide which route to go...

Answer 4

[CandiceChuDVM]

I contact the KOBAS 2.0 team to figure out why. They told me they haven't updated the system for a while, and let's why I got different numbers of pathways and genes. However, they will keep update it in the future.

Can anyone recommend other platforms for KEGG?

[WouterDeCoster]

I'm a fan of Enrichr: http://amp.pharm.mssm.edu/Enrichr/

This has also multiple other databases in addition to KEGG and GO.

[GenoMax]

Just curious. Since KEGG is no longer freely available how is Enrichr providing access to that data? The site says that they have 2016 data from KEGG.

[WouterDeCoster]

I hadn't thought about that and have no idea.

[CandiceChuDVM]

I tried Enrichr. However, it never displays my results. Does it support dog genome?
Can anyone see any results here: http://amp.pharm.mssm.edu/Enrichr/enrich?dataset=6jgx
I tried both chrome on Mac and IE on PC, and it won't work.

[WouterDeCoster]

Oh damn, I guess it's only for human and mouse. Another alternative is http://bioinfo.vanderbilt.edu/webgestalt/option.php But this one is not so well up to date.

[GenoMax]

Webgestalt does support dog genome though the annotations are much older as @Wouter already mentioned.

The current version of WebGestalt was updated on 1/30/2013. It supports eight organisms including human, mouse, rat, worm, fly, yeast, dog, and zebrafish.

[CandiceChuDVM]

I used gene symbol instead of gene id then it worked. However, the displayed results are only for human :(